Utilizing Artificial Intelligence (AI) and Image Recognition Technologies to achieve Potential Wireless Sensing of Surgical Resection Boundaries for Benign Prostatic Hyperplasia (BPH)

BACKGROUND AND OBJECTIVE: Precise intraoperative sensing is critical for optimizing surgical outcomes and patient safety. This study proposes an AI driven model, leveraging the nnU-Net architecture, to achieve real-time identification of surgical resection boundaries during transurethral resection of the prostate (TURP). While currently relying on image-based recognition, this work lays the foundation for integrating wireless sensing technologies, potentially transforming non-contact surgical guidance and biomedical applications. METHODS: In this research, we presented an nnU Net model capable of automatically predicting benign prostatic hyperplasia tissues, specifically encompassing the prostatic capsule, verumontanum, and bladder neck. The model was constructed using a contracting path and an expanding path, with Leaky ReLUs employed to fine-tune the learning rate. We conducted separate evaluations to assess the AI aided recognition performance for the prostatic capsule, verumontanum, and bladder neck. RESULTS: This study employed a 5-fold cross-validation approach to process each binary dataset. Specifically, for the prostate surgery capsule group, 228 images comprised the training set, while 58 images constituted the validation set. In the verumontanum group, 210 images were designated for training and 53 for validation. Similarly, in the bladder neck group, 236 samples were allocated to the training set and 59 to the validation set. The nnU-Net model was then trained and validated using these datasets, with its performance being assessed through the use of Dice coefficient, mean Intersection over Union (mIoU), mean Accuracy (mAcc), and overall Accuracy (aAcc) metrics.Prostate surgical capsule: mDice = 0.67, mIoU = 0.579, mAcc = 0.7, aAcc = 0.926;Verumontanum: mDice = 0.837, mIoU = 0.742, mAcc = 0.835, aAcc = 0.943;Bladder neck: mDice = 0.76, mIoU = 0.667, mAcc = 0.748, aAcc = 0.967These results demonstrate the model's performance in predicting and delineating these surgical boundaries.Conclusons Our findings highlight the potential of AI in real-time human sensing and pave the way for future integration with wireless technologies, enabling non-contact detection and precision healthcare solutions. This approach aligns with emerging trends in personalized biomedical sensing and wireless healthcare technologies.